As all outputs from secondary GPUs are in the same map as the ones from
the primary GPU, EglMultiBackend::findBackend always returned the
primary backend. To fix that, search in secondary backends first
This is to prevent hitting the case where there are no any outputs.
Ideally, it shouldn't matter in which order outputs are added or
removed, but the Workspace and AbstractClient subclasses don't work
with true headless mode. If there are no outputs, kwin can crash.
Due to the screen edges test not being an integration test, it's very
hard to change output related code in libkwin. screens.cpp needs to have
a few ifdefs to successfully compile.
This change rewrites the screen edges test as an integration test in
order to allow us using other components of kwin in screens.cpp and
screenedge.cpp without ifdef guards.
It's not a one-to-one port.
With the new output query algorithm, position of an output may not align
with its xinerama index. This can be problem on x11 as some protocols,
e.g. startup feedback, use xinerama indices to identify outputs.
Technically, the RandR specification doesn't say that CRTCs are stored
in the xinerama order, it only mentions that the first CRTC corresponds
to the primary output. However, such assumption was made by kwin prior
to output query changes, this change merely restores that behavior.
Active output is a window management concept. It indicates what output
new windows have to be placed on if they have no output hint. So
Workspace seems to be a better place for it than the Screens class, which
is obsolete.
This allows us to drop obsolete Screens class.
Currently, the Screens class is a thin wrapper around platform outputs +
some extra tracked state, e.g. active output, workspace geometry, etc.
This is a little helper that can be very convenient with our transition
from int-based screen ids to AbstractOutput.
As is, the main issue with int screen ids is that they are extremely
dynamic.
Instead of crashing because a nullptr property gets accessed, use
an explicit error message and ignore the offending object. This also
acts as documentation
With AbstractOutput being used more heavily, it makes sense to have
something like Screens::number() in the Platform class. As is, the steps
to get an output for a given point are awkward - first, get the screen
id, then use the screen id to get the output.
With the new virtual desktop model, we have an issue where the old
code that uses desktop() needs to be ported to desktops().
However, using no desktop() is better as we don't need to deal with
cases where a window can be on several desktops, which can be annoying
sometimes.
This change removes the desktop arg in electricBorderMaximizeGeometry()
and ports it to a Workspace::clientArea() overload that requires no
desktop.
Under the hood, Workspace::clientArea() still uses desktop(), but it
could also use a different strategy to compute the client area if the
window is on several virtual desktops, e.g. intersect client area on
every virtual desktop.
There's no need to guard the code, gbm and drm format definitions
are the same. Using GBM_BO_FORMAT_* even caused bugs, as that is
an enum and not a proper format identifier.
BUG: 441253
It's confusing to have two signals (virtualScreenGeometryChanged() and
screenGeometryChanged()) that indicate the same thing.
This change ports parts of kwin from the screenGeometryChanged() signal
to the virtualScreenGeometryChanged() signal with the main motivation to
drop the former.
The virtualScreenGeometryChanged() signal was chosen as the replacement
for the consistency sake with EffectsHandler's virtualScreenGeometry and
virtualScreenSize properties.
Currently, the EffectsHandler has two signals that are emitted when the
combined geometry of all outputs change - virtualScreenGeometryChanged()
and screenGeometryChanged(). Having two signals is most likely a
historical artifact.
This change untangles the screenGeometryChanged() signal from the
Workspace and makes it the same as the virtualScreenGeometryChanged()
signal.
Currently, there are a couple of issues with output querying on X11:
(a) if an output is changed, for example its transform has been changed,
then all outputs will be destroyed and created again
(b) it's possible to encounter the case where the platform has no
outputs. The X11Platform destroys all outputs, then queries new
outputs. The Workspace and AbstractClient sub-classes handle having
no outputs very poorly! It's even possible to hit a crash.
With this change, outputs will be queried similar to how it's done on
Wayland.
The current duration is 300ms, which is inappropriate for two reasons:
1. It's too slow
2. It doesn't match one of the standard durations. The effect should
ideally use those durations directly, but if this is not possible
for technical reasons, it should use the effective value of one of
those durations. Units.longDuration is 200ms which is the most
appropriate one to use here.
Therefore, change the animation duration from 300ms to 200ms to match
the standard Units.longDuration duration.
BUG: 441308
FIXED-IN: 5.23
This makes it easier to spot if the effect is activated while there is
only one maximized window.
Note that anchors.margins cannot be used unfortunately as it won't take
effect until the ExpoLayout item is fully constructed. It may take a
couple of cycles for the geometry to settle down, which makes reasoning
about how the effect works difficult.
BUG: 312146
Once an animation starts due to property change, it cannot be easily
re-targeted. This can cause undesired animation jumps if a property
changes during startup (for example from 0 to its final value).
With this change, the window heap's animation will be disabled until
the effect starts the intro animation.
The ExpoLayout delays relayouting for optimization purposes. However,
this means that new geometry will be available only after returning to
the event loop.
This change delays starting the intro animation so it can be started
with new geometries.
It's confusing to have globalPos() and geometry() as both can return the
same information.
This change drops globalPos() function as it's not used outside the
AbstractWaylandOutput class and renames setGlobalPos() to moveTo() to
avoid having a setter without matching getter.
According to the Qt documentation, if an item overrides the
geometryChanged() method, it must call the base class method.
QQuickItem::geometryChanged() is responsible for emitting signals such
as xChanged(), yChanged(), etc.
It's the same as the currentDesktop except that it's a VirtualDesktop
object.
The main advantage of the currentVirtualDesktop property is that it
doesn't change if a virtual desktop is added or removed in the middle.
It also makes dealing with virtual desktop changes easier.
With the new virtual desktops model, a window can be on several virtual
desktops. Currently, scripts have no any way to get the list of desktops
the window is on. This change addresses that.
Currently, window thumbnails may be placed behind the panel, which
doesn't look good.
With this, the window heap will cover the same area as the maximize
area.
This makes the WindowHeap more robust to layout changes. Currently, the
main issue with it is that thumbnails will be misplaced if the heap's
scene position changes.
Kirigami.ScenePosition automagically keeps track of the item's scene pos.
If a window is on several virtual desktops, Workspace::switchWindow()
will not work as expected because AbstractClient::desktop() returns the
id of the last virtual desktop.
Since the active window is most likely on the current virtual desktop,
we can use that for filtering out irrelevant windows.
The new overloads take the client (as context) and the desired screen id
or a point and return the client area.
The main motivation behind this change is to make the transition to the
new virtual desktop model where a window can be on several desktops less
painful.
In the recent refactor we made it so environment variables got synced to
the Application::processStartupEnvironment.
It then seemed safe and cleaner to remove the qputenv calls, but this
was not the case. It regressed the unit tests on CI and kglobalaccel
which is another path where kwin spawns clients.
Syncs the primary selection from wayland to X and from X to wayland.
Instead of doing it through the internal connection like the clipboard,
this sets/reacts to changes in SeatInterface::prrimarySelection directly.
BUG:422426
FIXED-IN:5.23
This effect is meant to be as a replacement for the present windows and
the desktop grid effect. It is written using QML.
So far, this effect implements only the basic features of the present
windows effect. Desktop management features will be added later.
CCBUG: 295775
CCBUG: 303438
This allows changing the type of desk to QVector<VirtualDesktop *>.
Based on the dont_activate flag, Workspace::sendClientToDesktop() will
try to focus the window if it's moved to the current virtual desktop.
In order to implement that, it needs to know whether the window has been
on the current desktop. c->isOnDesktop(desk) is a much sophisticated way
to do that.
Workspace::takeActivity() forces the modal dialog and its parent window
to be on the same virtual desktop. It uses int-based desktop ids for
that purpose.
This change ports the relevant parts of Workspace::takeActivity() to the
VirtualDesktop object.
This way if kwin_wayland crashes we don't need to ensure that new
environment variables need to get synced across the new env.
This fixes an issue where spawning an xwayland application from a
wayland window that survives a crash would fail.
By moving the logic here we no longer need to wait for kwin_wayland to
start before starting plasmashell or even ksmserver as all environment
variables are set. As long as the wrapper is ready we can continue
starting and clients will just block on connect.
That should still allow for both a lot of optimisations both for speed
and cleaning up the startplasma-wayland scripts.
This will be addressed in follow up patches.
Use of kwin_wayland directly with xwayland is still supported for
testing.
The initial goal of kwin_wrapper was that it would set up all relevant
environment variables that will be used by the clients of kwin.
This meant having to cache old variables before we overrode them and
pass additional settings to the backends for the old wayland_display. It
works, but with X11 about to move too ends up being unnecessarily
complex. In hindsight it's easier for kwin to have the environment
represent the platform it is currently on, and have kwin explicitly set
variables in the QProcessEnvironment of the session it forks itnto.
This patch is changed so that we set the wayland socket name used by the
wrapper explicitly which is then used by the process environment kwin
uses for the main session.
If a window is on several desktops, AbstractClient::desktop() will
return the id of the last desktop.
For example, if a window is on virtual desktops A and B, the desktop()
function will return the id of desktop B. This can be the culprit for
bugs such as window snapping not working as expected when moving a
window on virtual desktop A, e.g.
- moved window is on desktops A, and C. desktop() returns the id of C
- snap candidate (l) is on desktops A, and B. desktop() returns the id
of B
Even though the snap candidate window and the moved window are both on
the same desktop (A), the moved window won't be snapped because the
desktop() function returns garbage values.
To fix that, the workspace needs to check whether the window is on the
current desktop.
For what it's worth, that's also how the workspace handles windows being
on multiple activities.
This patch has one behavioral change - raiseOrLowerClient() will not
work if the client is not on the current virtual desktop.
However, raiseOrLowerClient() can be called only in two cases:
* user triggers the raise or lower shortcut for the active client. Since
the active client is on the current virtual desktop, it's not an issue
* an x11 window restacks itself. It makes no sense if an x11 window
restacks itself while it's inactive or not on current virtual desktop.
Also, the Opposite restack mode is rarely used, some window managers
don't even bother implementing it. So, having such a constraint should
not be a problem.
The main reason for not allowing raiseOrLowerClient() for windows that
are not on the current virtual desktop is that a window can be on
multiple virtual desktops. If a window is on A and B virtual desktops,
the only logical option is to toggle stacking position if the window is
on the current desktop. It's the only viable option as kwin does not
maintain per virtual desktop stacking order.
IN_FORMATS contains information about which buffer formats can be
scanned out with a given drm plane. Using these plane-specific
format+modifier combinations as well as explicit modifiers in
general can yield bandwidth and performance improvements,
especially in multi-gpu systems.
We use surfaceless contexts with internal windows. We also require
the EGL_KHR_surfaceless_context extension for making context current
without outputs.
Arguably, we could use pbuffers, but since mainstream drivers (Mesa and
NVIDIA) support surfaceless contexts, the extra complexity doesn't buy
us anything.
This reverts commit ac16bef409.
It causes crashes and color channels seem to be swapped.
GLTexture::toImage() needs more work before it can be used in the
screenshot effect, or maybe dropped.
In a recent refactor screens are only created when our xdg_surface has
been acked. This leaves a window where m_waylandCursor is null and
events are still being processed.
With per-screen rendering, every output may have different EGLConfig.
Having a single global EGLConfig doesn't work out well.
This change removes Platform::sceneEglConfig(). It's used primarily to
create the global share context. In hindsight, the global share context
can be created without EGLConfig as it's never made current.
EGL_NO_CONFIG_KHR is part of EGL_KHR_no_config_context extension, which
is supported by both Mesa and NVIDIA driver so it should be safe to make
it mandatory.
At this point, desktop() is -1 since the client hasn't been moved to any
virtual desktop yet.
It seems like checking desktop() is a typo. It makes more sense to check
desktopId.
This ports relevant apis in WindowRules to the VirtualDesktop class.
If the client has no desktop rule, the desktop list that has been passed
to the checkDesktops() function will be returned.
If the client has a desktop window rule, the checkDesktop() function
will return a list with a single VirtualDesktop object or none if the
window is forced to be on all virtual desktops.
The main motivation behind this change is to reduce the number of usages
of int-based virtual desktop apis. int-based desktop ids are highly
unreliable. For example, if a new virtual desktop is inserted in the
middle or removed in the middle, the desktop ids will change. This makes
working with virtual desktops code more challenging due to its behavior.
This is not an issue with VirtualDesktop objects.
This change avoids unnecessary copying of QImage data when updating
textures using QImages that have non-ARGB32_Premultiplied formats.
There are three cases where texture unpacking can not be done/is
not worth doing: if GLES is used, and the extension for unpacking
subimages is not supported; if the entire image is being updated;
or if a format conversion needs to be done, in which case only the
region to update will have its format converted, and the resulting
image can be passed directly to glTexSubImage2d without setting
unpacking parameters.
These repaints were added to fix visual artifacts that appear when
shadow is removed, but since items schedule repaints when needed, we can
remove these two.
Wayland- and X11-specific code paths are not needed.
If the window is on all virtual desktops, Toplevel::desktops() will be
an empty list.
If the window is on a particular virtual desktop, Toplevel::desktops()
will contain that desktop. On X11, the desktops list can have at most
only one item.
The old calculation ignored the render time and safety margin of the
frame scheduling algorithm, which makes VRR activate properly only when
the refresh rate is much lower than the maximum refresh rate.
This further decouples scene items from scene windows. The SurfaceItem
still needs to access the underlying window, I would like to re-iterate
over that later.
With this change, it will be possible to introduce WindowItem factory
function in the Toplevel class.
toplevel.h is included in many places. Changing virtualdesktops.h may
trigger rebuild of all kwin.
With this change, only cpp files that use virtualdesktops.h will need to
be recompiled.
The main idea behind _NET_WM_FRAME_OVERLAP is to extend the borders of
the server-side decoration so the application can draw on top of it. It
was inspired by similar feature in Windows.
However, _NET_WM_FRAME_OVERLAP is basically unused. Neither GTK nor Qt
support it and I have never seen any application that uses it.
At the moment, kwin is the only compositing window manager that supports
_NET_WM_FRAME_OVERLAP. Neither mutter nor compiz nor compton and so on
support it.
Since _NET_WM_FRAME_OVERLAP is practically unused, there's no point for
keeping supporting it.
This change shouldn't affect any existing app as _NET_WM_FRAME_OVERLAP
atom is not listed in _NET_SUPPORTED.
This ensures that surface role objects can be still accessed when either
XdgToplevelClient or XdgPopupClient is destroyed. As far as I know, it
doesn't cause any problems right now, but this makes resource destruction
handling consistent, we tend to perform cleanup in aboutToBeDestroyed
signal slots.
SceneOpenGLShadow::prepareBackend used to use QImage::Format_Indexed8,
plus a special code path in GLTexture, to create single-channel OpenGL textures.
Now that Qt supports QImage::Format_Alpha8, this workaround can be removed.
Currently, a vertex coordinate is transformed first, then mapped to
the global screen coordinates. This causes a problem if a transform
is applied to the top-most item and child items are not at (0, 0).
For example, scaled windows may have popping out sub-surfaces, etc.
With this change, the item transforms will be computed differently. For
example, if the parent item is transformed, a child's transform will
look as follows
[Parent's translation][Parent's transform][Child's translation]
instead of
[Parent's translation][Child's translation][Parent's transform]
In the future, I'd like to get rid of the Item::setTransform() call in
OpenGLWindow::performPaint() and have either AnimationEffect or
libkwineffects call Item::setTransform().
BUG: 440201
Deleting QQuickWindow will force any remaining runnables to complete.
This is important for any items (such as our new ThumbnailItem) that
deletes resources in queued cleanups.
We need to invoke this manually whilst we still have an active render
control with a valid context, otherwise remaining runnables will no-op.
We also then want to manually delete our scene contents before the view
so they're not orphaned temporarily.
This fixes a memory leak observed with the new Overview effect.
If a sub-surface should be considered unmapped, make the corresponding
item invisible.
Note that this must apply only to sub-surfaces. Whether the main surface
should be visible is up to kwin. It may want to play an animation when
the toplevel surface is unmapped, etc.
In GLTexture::update, reuse the mechanism from the GLTexture(QImage, ...)
constructor to upload QImage data directly when the QImage format
has compatible GL format and type codes. Before this change, all QImage
formats other than ARGB32_Premultiplied required a format conversion.
As per the xdg-shell specification, the compositor can only attach
a wl_buffer to an xdg_surface once the compositor has sent an
xdg_surface.configure event. Previously, there was a race condition
in which WaylandOutput objects were added/enabled on creation, making
it possible for the EglWaylandOutputs to be set up and for
EglWaylandBackend to draw a frame (thereby attaching a buffer), all
before the xdg_surface.configure configure event was received.
Compositors other than kwin would then send an unconfigured buffer
error and close the connection.
This commit avoids the problem by not marking the WaylandBackend
ready for use until all the output surfaces constructed at startup
have received their initial configure event. (In practice, this
takes one additional roundtrip.)
We're porting away from using plugin keywords (and have done that for
the breeze and oxygen decoration). However to allow that we need to
adjust the loading code.
Keep the old codepath for compatibility.
BUG: 440786
Blending is quite expensive especially with software rendering.
In the case of Firefox on Wayland, it uses a ARGB8888 buffer but marks the
entire surface as opaque, so the alpha channel can be ignored.
CCBUG: 440386
When the last output gets disconnected, create a virtual output as a
placeholder until we have access to a physical output again. While
this placecholder never gets rendered to, with virtual outputs in
general that is possible (with gbm and qpainter atm) and can be done
for future use cases like wireless displays.
CCBUG: 420160
CCBUG: 438839
QImage is implicitly shared. However, methods such as QImage::bits() may
detach. In order to avoid the detach, one either needs make the image
const or use QImage::constBits() method.
Currently, the GLTexture class uses QImage::bits() all over the place.
It prevents implicit detaching by storing images in const variables. Such
code is fragile. Adding an intermediate (optional) conversion step may
result in odd performance regressions.
Currently, the shape region is not copied to the Deleted. If either
SurfaceItemX11 or SurfaceItemXwayland needs to build quads for a shaped
window, it won't be able to do so properly because the corresponding
x11 window is long time gone.
If the shape region changes before the window is unmapped, you may still
see visual artifacts. Unfortunately, the only way to fix that bug is to
switch to wayland.
BUG: 440001
BUG: 438458
BUG: 435378
https://invent.kde.org/plasma/kwin/-/merge_requests/1220
This reverts commit 44ad9437c4 because it
breaks build:
platforms/x11/common/eglonxbackend.cpp:224:66: error: cannot initialize a parameter of type 'EGLNativeWindowType' (aka 'unsigned long') with an rvalue of type 'xcb_window_t *' (aka 'unsigned int *')
surface = eglCreateWindowSurface(eglDisplay(), config(), &window, nullptr);
^~~~~~~
It is perfectly valid to have a case where an item exists but has no
window. During these times the item will never be rendered.
Rather than guard in updateOffscreenTexture, it's more economical to
disable the frame rendering in the first place. Especially as then it's
easy to extend to item visibility.
BUG: 440318
Makes it possible to apply the dpms settings per screen instead of
applying it to all of them, which is wrong at many levels.
Will be even more important with other effects like rotation.
If kwin is explicitly restarted (kwin_wayland --replace) we reset the
crash count so that you can run it more than 10 times without exiting to
the logout screen.
Resetting to 0 causes a issue as that triggers ksplash to be shown
again.
BUG: 434489
We inintially created the wrapper to be light so avoided the QtCore
dependency and did it all in C.
In hindsight changing requirements means we want to add several more
features in here. The first step is to port to a more readable language
and toolkit.
Currently, the entire screen will be repainted when using the qpainter
render backend. With this change, kwin will repaint only the dirty parts
on the screen.
A better alternative is to return the damage region in the beginFrame()
function. This way, the render backend can force full screen repaints or
indicate what parts of the buffer needs to be repainted (for buffer age)
The main motivation behind the split is to simplify client buffer code
and allow adding new features easier, for example referencing the shm
pool when a shm buffer is destroyed, or monitoring for readable linux
dmabuf file descriptors, etc.
Also, a referenced ClientBuffer cannot be destroyed, unlike the old
BufferInterface.
If some popups of active window present, there was glitches on switching
such as stale popups or the popup happened below window.
Dismissing all the popups solves the problem.
BUG: 431046
If window quads need to be generated after the wl_surface is destroyed,
the SurfaceItemWayland::mapToBuffer() function will return wrong values.
In order to fix that, we need to store the last surface-to-buffer matrix
in SurfaceItem.
These appear to be remnants from a time where the current tabs were individual KCMs.
Remove some unneeded stuff and convert the only relevant registration to the normal K_PLUGIN_FACTORY macro
Since we adapted inputmethod to support methods like ibus, the input
method can be active but not have a visible panel.
This includes an extra property that will indicate us if the panel is
visible at any time. This will allow us to properly render the virtual
keyboard hide button in Plasma Mobile (or wherever we need it).
So far calling setActive(true) would issue a deactivation then another
activation. This sometimes makes maliit crash and we can achieve the
same result just by just issuing a reset.
Xwayland windows have two geometry types that are used during rendering
- wl_surface's geometry and x11 window's geometry. Unfortunately, it's
not possible just ignore the x11 window geometry because the window may
have a custom shape region. Pixels outside the shape region can have
arbitrary values; it's not guaranteed that the area outside the shape
region will be transparent.
In some cases, the x11 window geometry can be larger than the wl_surface
rect. That breaks repaint logic as the SurfaceItemXwayland can draw
outside the bounding rect.
Specifically, this issue can be seen while resizing Steam interactively.
Another example is where a video game enters fullscreen mode. I
occasionally see that Red Dead Redemption II window leaves a ghost after
itself when it switches to fullscreen mode.
It's not a perfect solution, but on the other hand, it doesn't look like
there's a better way to handle this due to the fact that there are two
conflicting geometry sources.
If a QObject is exposed to js using QJSEngine::newQMetaObject(), a new
instance of it can be made only with constructors exposed by Q_INVOKABLE
At the moment, QTimer's constructor is not Q_INVOKABLE, so code such as
`const timer = new QTimer()` will not work.
BUG: 439630
FIXED-IN: 5.22.4
If the client had created an xdg-toplevel-decoration and has called the
set_mode() request with csd before the initial configure event is sent,
we still need to send an xdg-toplevel-decoration configure event with
csd mode.
If the output transform changes, the geometry may change as well;
otherwise qpa may not call QWindowSystemInterface::handleScreenGeometryChange()
with the corresponding platform screen.
If the visiblityChanged signal is emitted and then the window is
immediately destroyed, the slot will still be executed because the
connection has type of Qt::QueuedConnection.
DrmPipeline is now what contains all the thing related to drm calls,
instead of DrmOutput. This allows for some more flexibility and tidies
the code up a bit. Additionally instead of rolling back changes if
presentation fails, changes are directly tested with atomic test only
commits.
In order to not make misdetections with sub-pixel movements the gesture
recognizer completely ignored those. This fails however when you move
your fingers very slowly, so instead accumulate the delta and only use
it for direction detection if it's high enough.
According to the spec, if the wl_subsurface.place_below request is
called with the parent surface, the sub-surface must be placed below the
parent surface.
BUG: 438808
While always calling showCursor isn't a problem for when there's no
pointing device as the cursor image is empty in that case, it can
cause a temporarily stuck cursor image when it's supposed to be
hidden because of touch input
Originally, only x, y, width, and height property setters were used. A
bit later "position" and "size" helper setters were added to simplify
code in SurfaceItem sub-classes. It seems like setX(), setY(), setWidth(),
and setHeight() are unused now, so remove them.
The tablet protocol offers us the cursor in a surface. This patch makes
sure we connect to the surface so that we can also refresh the
representation if the client submits a change.
BUG: 437777
The XDamageDestroy has a weird requirement saying that it can be called
as long as the X11 window is still valid. On the other hand, one could
argue that it is more intuitive if the damage handle becomes inert if the
associated window is destroyed. Unfortunately, that's not the case and
as git history shows, it's an easy way to shoot yourself in the foot, we
had the exact warning many years ago.
The problem with the XDamageDestroy API is that it is simply unreliable
given the asynchronous nature of communication between kwin and xorg.
Anyway, with X11 sunsetting, let's destroy the damage handle only when
the X11 window is unmapped and not bother too much about it.
If a wl_eglstream buffer is attached to a surface, but then later a different
type of buffer, such as a dmabuf, is attached to the same surface, kwin will
mistakenly keep trying to acquire frames from the EGLStream associated with the
previous buffer. This can happen if an Xwayland window is made full-screen
causing it to switch to the flipping presentation path, for instance. The
result is that the window contents will no longer be updated.
Instead, the eglstream backend's loadTexture and updateTexture functions should
first pass the buffer to eglCreateStreamAttribNV. If it fails with
EGL_BAD_STREAM_KHR, that indicates it is indeed a wl_eglstream, and that we've
already associated a server-side EGLStream with it in attachStreamConsumer, so
we can proceed as usual. If it fails with EGL_BAD_ACCESS, though, that
indicates it is not a wl_eglstream and we should fall back to the parent class
which handles attaching other buffer types. If it doesn't fail at all, that
means the client tried to attach a new wl_eglstream to a surface without first
attaching the stream consumer. There's not really a great way to handle this,
so just re-use the previous EGLStream.
This is to ensure that both kwin and Qt use the same EGLDisplay. Note
that the native context handle can have no display, however it's very
unlikely to happen.
There are EGL implementations that will refuse to create a context if
the share context belongs to other EGLDisplay. Currently, that's the
case on some platforms.
If eglGetPlatformDisplayEXT() is called with the same parameters, it'll
return the same EGLDisplay. The main motivation behind this change is to
ensure that both kwin and Qt share the same EGLDisplay, QtXCB calls
eglGetPlatformDisplayEXT() without any additional args.
If EGL_PLATFORM_X11_SCREEN_EXT is not specified via attrs, the EGL
implementation will use the default screen, which is the same as
m_x11ScreenNumber.
The decorationPressId never gets reset as the decoration isn't below the
touch point anymore after the quick tile, so if we have no decoration
under the touch point reset the decorationPressId to -1.
CCBUG: 430560
If compositing is disabled, compositor will be null.
We hit a path where we could destroyOffscreenTexture when the
compositing is disabled then again when the Thumbnail is destroyed. In
this case we query Compositor::self()->scene() and crash.
Currently, thumbnail items are rendered by kwin. This means that qtquick
code cannot do things such as applying shader effects to window thumbnails
or simply draw custom controls on top of thumbnails.
With this change, task switchers and qml extensions will be able to
place their own contents on top of thumbnails and apply custom effects
to them.
In order to integrate window thumbnails, a window is rendered on kwin
side using its own opengl context. A fence is inserted in the command
stream to ensure that the qtquick machinery doesn't start using the
offscreen texture while there are still rendering commands being executed.
Thumbnails are rendered into offscreen textures as we don't have full
control over when qtquick windows render their contents and to work around
the fact that things such as VAOs can't be shared across OpenGL contexts.
WindowThumbnailItem and DesktopThumbnailItem act as texture providers.
It's unused and the advantages of keeping it are outweighed by the
disadvantages - the returned value is dependant on the window type.
If you need to draw a drop-shadow that matches the shape of the window
or something along that line, render the window into an offscreen
texture and sample the alpha channel in a fragment shader.
Putting the OpenGL post processing rotation into its own class cleans
the EglGbmBackend code up a bit and adds post processing rotation for
the EglStreamBackend
It is error-prone to have multiple sources for the same data. If the
base implementation (Compositor::compositing()) changes, other helpers
can get out of sync.
When finishing compositing, Workspace::compositing() will return true,
but it will be preferred if it returns false instead so kwin can properly
update x11 window visibility status, etc.
Instead of checking whether the compositor has a scene, check the status
of the compositor. It will not be "On" during teardown.
This allows creating a GLTexture object and attaching a dmabuf to it.
Currently, we can do that by using the foreign GLTexture constructor,
but it makes the deletion of the texture handle more error prone.
In the future, we can add a method that allocates the texture storage,
but there's no need for that yet.
Currently, if discard() is called, kwin will crash because
EglPixmapTexture does not override the discard method.
In principle, neither GlxPixmapTexture nor EglPixmapTexture should mess
around with internals of the GLTexture class. It is better to have a
wrapper texture with a bind method, which will re-bind the pixmap to the
opengl texture if necessary.
It seems like without a surface creation of the scene fails somehow. At
least until the exact problem is solved, update outputs for EglStream
gpus before creating the EglStreamBackend.
BUG: 438363
The pointing finger cursor is used to show that a piece of text is a
clickable link, and it is inappropriate to use it in other contexts.
Regular UI elements that do something when single-clicked continue to
use the standard arrow cursor in other contexts, so let's follow that
convention in the Present Windows and Desktop Grid effects too.
BUG: 421928
FIXED-IN: 5.23
This commit removes the separators lines below "scale method" and
"rendering backend", options here shouldn't be separated since they
are all related to eachother.
Currently, each effect's list items expands when you click on any
non-interactive part of it, but it does not collapse when you click on
it again.
With this commit, now it does.
BUG: 421883
FIXED-IN: 5.23
Makes for an odd decoration.
We generally will not have an app_id in the cases where we are creating
a token just in case it's necessary down the line.
If a client already has powers to move clients around, let them also
activate using the right protocols as well.
This is useful to be able to implement SNI or Notifications where the
shell doesn't get full surface activation when interacting with it.
This hack was added in the past to prevent the combobox from being
pushed out of view when the label next to it was very long, as it can be
in certain languages. However it had the effect of making the label get
elided in *all* languages, and also the issue it was working around is
no longer present; even with very long text, the combobox no longer gets
pushed out of view, and instead the formlayout first switches to mobile
view and then elides the Label, exactly as intended. So we can safely
remove the hack.
BUG: 438504
FIXED-IN: 5.22.2
If a redirected animation reaches the end, the timeline value will be 0,
i.e. the interpolated() function has to return the from value, not `to`.
BUG: 438368
When the aboutToDestroy signal is emitted, the compositor object is
already partially destroyed. This contradicts to the name of the signal.
In addition to that, it will be better to call the stop method in the
destructors of X11Compositor and WaylandClient as it allows them to add
custom cleanup code when compositing is turned off.
If a window is animated using DeformEffect, allocate an offscreen
texture with the scale factor of the screen where the window is
considered to be on.
Also, resize the offscreen texture on demand in order to make resource
handling more efficient.
At the moment, we handle window quads inefficiently. Window quads from
all items are merged into a single list just to be broken up again.
This change removes window quads from libkwineffects. This allows us to
handle window quads efficiently. Furthermore, we could optimize methods
such as WindowVertex::left() and so on. KWin spends reasonable amount
of time in those methods when many windows have to be composited.
It's a necessary prerequisite for making wl_surface painting code role
agnostic.
Window quads need to be in some coordinate space. Since we want items to
be used not only for rendering windows, window-local coordinates do not
suffice.
This change makes scene items generate quads in the item-local coordinate
space.
The model matrix is used to map the quads to the global screen coordinate
system from the item coordinate space.
Since the quads are in the item local coordinate space, the mvp matrix
needs to be updated on every draw call. Given the render data, tracking
the last mvp matrix won't result in less glUniform calls. If this indeed
becomes a serious performance bottleneck, we can explore the possibility
of dumping mvp matrices in a UBO, which have been introduced in OpenGL
3.1.
The scene items depend on the scene windows for caching window quads.
The goal of this change is to move window quads management to item.
Merging window quads in one list and then splitting them is inefficient,
it will be highly desirable if window quads are removed from the public
api so we can optimize window quad management.
With this change, the window quad type becomes irrelevant to render
backends for the most part. Note that the Xrender backend is a bit
nitpicky about window quads, so the shadow item doesn't create generic
"WindowQuadShadow" quads anymore.
Drop-shadows with the software render backend impact performance quite
significantly. It also makes it easier to prepare render backends for the
item based design.
Use FontMetrics to calculate the expected max text width so it
doesn't overflow with wider fonts, specially in some languages
like pt_BR
BUG: 438193
FIXED-IN: 5.22.1
Offers shoul be set on a source before data_device.set_selection. Doing
so afterwards appears to be against the spec and as such we may not pass
the update to wayland clients / klipper.
So far, we were creating a model view with the complete scene rendered
(even if we didn't render the windows themselves). This required us to
have a big glPerspective spanning the entire scene and we were just
cropping it as we rendered it into a smaller texture.
This changes our scenes so we have the correct matrix set up at all
times.
Specifically in the case of the Pinephone, this solves the following
issue where we were unable to connect external displays because it
exceeded GL_MAX_VIEWPORT_DIMS:
https://invent.kde.org/teams/plasma-mobile/issues/-/issues/11
When the panel disappears, just reconsider the panel's state but don't
just stop sending updates to the input method.
Some input methods are just helpers that show and hide as necessary.
The Xrender backend was added at the time when OpenGL drivers were not
particularly stable. Nowadays though, it's a totally different situation.
The OpenGL render backend has been the default one for many years. It's
quite stable, and it allows implementing many advanced features that
other render backends don't.
Many features are not tested with it during the development cycle; the
only time when it is noticed is when changes in other parts of kwin break
the build in the xrender backend. Effectively, the xrender backend is
unmaintained nowadays.
Given that the xrender backend is effectively unmaintained and our focus
being shifted towards wayland, this change drops the xrender backend in
favor of the opengl backend.
Besides being de-facto unmaintained, another issue is that QtQuick does
not support and most likely will never support the Xrender API. This
poses a problem as we want thumbnail items to be natively integrated in
the qtquick scene graph.
With the ongoing scene redesign, it needs to be rewritten. However,
given that it is not used widely based on support information from
various bug reports and our available man power is sparse, the most
reasonable thing is to drop the effect, unfortunately.
With the ongoing scene redesign, it needs to be rewritten. However,
given that it is not used widely based on support information from
various bug reports and our available man power is sparse, the most
reasonable thing is to drop the effect, unfortunately.
With the ongoing scene redesign, it needs to be rewritten. However,
given that it is not used widely based on support information from
various bug reports and our available man power is sparse, the most
reasonable thing is to drop the effect, unfortunately.
With the ongoing scene redesign, it needs to be rewritten. However,
given that it is not used widely based on support information from
various bug reports and our available man power is sparse, the most
reasonable thing is to drop the effect, unfortunately.
static cuts[] array was initialized only once with copies of non-const
objects, so when new shortcuts are configured, the old objects were still
referenced.
Fixed by having non-static array instead, so actual objects are copied
on the init every time.
BUG: 359141
Some static code analysis tools break when analysing kwinbindings.cpp.
This change moves the contents of kwinbindings.cpp file in useractions.cpp
to make those tools happier. It also makes code more readable.
The KCM now manages the RuleSettings config objects directly,
instead of using a list of `Rules` objects as an intermediary.
This highly reduces the overhead, improving loading and saving
times, and also enables a better use of KConfig capabilites.
The config state is now automatically tracked by KConfigXT
objects. Whenever the user edits either the rule list, or any
specific property within a rule, the `needsSave` state is
updated accordingly.
BUG: 421564
FIXED-IN: 5.23
Previously, the only way to access the rules list was via
the `rules()` and `setRules()` methods, so the actual
settings objects were not accesible.
This commit adds methods to retrieve, insert, remove or reorder
the rules within the list.
Since every individual rule is stored as a KConfig group, and
they are not designed to be dynamically renamed, using consecutive
numbered groups and store only the total count is problematic.
So we also add a new stringlist setting to store the rules group
names and their order.
Now any group name is valid. To avoid collisions use random QUuids
as group names for newly created rules.
Even though there is a clientArea() overload that takes const abstract
clients, QJSEngine seems to be unable to find the right clientArea()
overload if a mutable abstract client has been passed.
BUG: 437507
Currently, the frameRendered() signal is emitted every time an effect
calls paintScreen(). This means that the frameRendered() signal can be
emitted more than once when effects such as slide are active. However,
we'd like if it's emitted only once before buffers are swapped.
This allows different input methods to get information about what's
beign typed from the actual hardware. This is especially useful for
non-latin script languages.
If the user has chosen to have a virtual keyboard (i.e. prefered maliit
over none in the KCM) they should get it. If it's too annoying it should
either be disabled or we can fix it so it gets less in the way.
This is especially important since some hardware registers itself as a
keyboard even if it's not a fully functional keyboard.
EffectWindow::shape() doesn't fit the item based design. This change
ports the blur effect away from the shape() function to the rect()
function. The XShape extension was introduced when windows with an alpha
channel wasn't really a thing. Setting a shape served as a way to clip a
window, the most notable example is xeyes.
If an application relies on the xshape extension to actually clip the
window and not to force the window manager not to put the server-side
deco, it most likely doesn't support translucency and therefore it
shouldn't set the blur region to begin with.
This change makes the blur effect ignore the xshape region similar to
the background contrast effect. It allows us to decouple a bit more
effects from the rest of rendering machinery and thus make it easier for
us to move forward with the scene redesign goal.
We used to get a weird line around the window decoration because the
clamping would hit the outside of the rendered decoration.
To make sure we fall inside take the displayPixelRatio into account.
Also on non-integer scales, make sure we are actually falling inside.
It was needed to work around visual glitches in the wobbly windows
effect. Since the wobbly windows effect renders the animated window into
an offscreen texture, we don't need this workaround anymore.
Furthermore, rather than using half-pixel correction, it is more
desirable to use an offscreen texture as it results in simpler design.
Performance-wise, it's not that bad that we need to start looking for
other ways to get rid of the seams between window contents and deco.
DeformEffect is the base class for effects that transform the window
quad grid, e.g. wobbly windows or magic lamp. The main difference
between normal effects and offscreen effects is that the latter renders
animated windows into offscreen textures which are later deformed.
The offscreen texture approach is superior to the half-pixel correction
approach as it produces more visually appealing results. Even with
half-pixel correction, you're still going to see jagged lines between
the main surface and the server-side decoration or drop-shadow.
It is also needed to reduce the number of usages of WindowPaintData::quads
which is needed to move forward with the scene redesign goal.
If the window borders change in such a way that the buffer geometry
doesn't change, but the client geometry does change, we need to configure
the wrapper window and maybe the client window so there are no black
borders, etc.
Currently, there is an assumption in the moveResizeInternal() method
that if the buffer geometry changes, then the client geometry will also
change. However, as it turns out, it's not completely true. For example,
if the window is maximized while its geometry already matches the
maximize area, the buffer geometry won't change, but the client geometry
will change.
This change adds missing geometry checks. *geometryBeforeBlocking
corresponds to the old geometry. We can use it to decide if the geometry
has changed.
If the geometry window rule must be forced, the move resize method has
to force it, but similar to size constraints, we assume that the caller
of the move resize function will do the right thing. So, remove the
debug message to be consistent with other client types.
Similar to the slide effect, the desktop grid can use the clip region to
clip windows when needed. This will improve performance because the desktop
grid effect will do less work. Currently, it clips quads on its own in
prePaintWindow, copies them over in a temporary list, and lets the opengl
scene clip the window quads again.
We can clip windows differently by just passing a custom region to the
paintWindow() function down the effects chain, which the desktop grid
effect already does.
The old behavior and the old bugs are preserved. The Xrender code path is
unaffected.
In general, on Wayland, there is no such a thing as "window scale factor"
because sub-surfaces can have different buffer scales. However, we know
the scale factor of the output where the window is considered to be on. So
we can use the screen's scale factor as the window's scale factor. In
most cases, it will produce the correct result.
For the scene redesign, window quads need to be removed from the effects
api. This change ports the screenshot effect away from quads.
In order to exclude the server-side decoration from the window
screenshot, the effect will render the contents of the client geometry
into the offscreen texture. If the window is client-side decorated, the
entire window will be rendered as before.
The good thing about this approach is that the screenshot effect will do
less work, it won't loop over window quads to compute the bounding rect
or filter out the deco quads.
The backend can now optionally wait for the scene to be created before
it updates its outputs, which is necessary for better atomic tests in
the DRM backend.
The SurfaceItemXwayland has to re-build quads if the shape changes.
The SurfaceItemX11 doesn't need a similar fix as it already discards
quads whenever the shape region changes.
The rnew name is applied when the user hits the return or enter key to
finish renaming, but this is an invisible UI and the user might not
think to do that. This commit replaces the "Rename" button with a
"Confirm new name" button while the text field is editable, so that it's
visibly clear how to finish renaming.
Similar to the feature to allow inline renaming in file managers, you
can now trigger renaming for a desktop by double-clicking on its list
item.
BUG: 421882
FIXED-IN: 5.23
Currently, the implementation of the DecoratedClient and the decoration
renderer are strongly coupled. This poses a problem with the item based
design as the ultimate goal is to have scene items construct paint nodes
which are then fed to the renderer. The DecorationItem has to have
control over the decoration texture. Another issue is that the scene
cannot smoothly cross-fade between two window states if the decoration
is removed, e.g. from fullscreen mode to normal and vice versa.
This change moves the decoration renderer to the decoration item. With
the introduction of a generic scene texture atlas, we hope to get rid of
the decoration renderer altogether.
AbstractClient::doPerformInteractiveMoveResize() is only used by the
X11Client to reset a boolean flag when the client doesn't support sync
counters.
X11Client can call performInteractiveMoveResize() only in two cases: the
sync request timer expires or the client increments the sync counter.
This change removes the AbstractClient::doPerformInteractiveMoveResize()
function and adds a function to handle the case where the sync timer
expires explicitly. This removes a virtual function in the AbstractClient
and makes code more readable.
Window management features were written with synchronous geometry
updates in mind. Currently, this poses a big problem on Wayland because
geometry updates are done in asynchronous fashion there.
At the moment, geometry is updated in a so called pseudo-asynchronous
fashion, meaning that the frame geometry will be reset to the old value
once geometry updates are unblocked. The main drawback of this approach
is that it is too error prone, the data flow is hard to comprehend, etc.
It is worth noting that there is already a machinery to perform async
geometry which is used during interactive move/resize operations.
This change extends the move/resize geometry usage beyond interactive
move/resize to make asynchronous geometry updates less error prone and
easier to comprehend.
With the proposed solution, all geometry updates must be done on the
move/resize geometry first. After that, the new geometry is passed on to
the Client-specific implementation of moveResizeInternal().
To be more specific, the frameGeometry() returns the current frame
geometry, it is primarily useful only to the scene. If you want to move
or resize a window, you need to use moveResizeGeometry() because it
corresponds to the last requested frame geometry.
It is worth noting that the moveResizeGeometry() returns the desired
bounding geometry. The client may commit the xdg_toplevel surface with a
slightly smaller window geometry, for example to enforce a specific
aspect ratio. The client is not allowed to resize beyond the size as
indicated in moveResizeGeometry().
The data flow is very simple: moveResize() updates the move/resize
geometry and calls the client-specific implementation of the
moveResizeInternal() method. Based on whether a configure event is
needed, moveResizeInternal() will update the frameGeometry() either
immediately or after the client commits a new buffer.
Unfortunately, both the compositor and xdg-shell clients try to update
the window geometry. It means that it's possible to have conflicts
between the two. With this change, the compositor's move resize geometry
will be synced only if there are no pending configure events, meaning
that the user doesn't try to resize the window.
We have 2 equivalent code paths for x11 and wayland unnecessarily, unify
them under the same method.
Rename m_clients to m_x11Clients so that it's clear what the difference
is between m_clients and m_allClients.
If textFromValue is set, valueFromText must be.
A lot of other implemenetations were using parseInt() which fails at
reading "3,000" due to the comma. Qt's fromLocaleString handles that but
needs any text stripping.
This implementation isn't perfect, but really we need to work on a good
upstream solution.
BUG: 437472
kwayland-server was updated to follow the result of whether the data
offer was accepted before calling drop.
Our bridge was not updated.
We don't have enough data to know which mimeType the XWayland client
actually has accepted, as that doesn't exist at an X11 level, but it
doesn't have a big actual impact. Despite which mime type gets accepted
any mimetype can be retrieved in the drop event.
Tested against "dropsite" from QtBase widget examples forced into xcb
mode using dolphin as a source.
BUG: 437406
attachStreamConsumer and resetOutput are not called as part of the scene
render and as such the context might not be set.
All credit goes to Simon Spinner for his investigation.
BUG: 437573
Ensures that m_active and the input context do not get out of sync and
that the right signals are always emitted by always using setActive to
activate.
Uses the context to define InputMethod::isActive.
Detach from the serials that the client send us, since they don't really
matter to the input method and track our own.
Also issue a commit state whenever we adopt a new input context. This
can happen for reasons that don't come from the client, like for example
the input method has just been enabled.
In practice, this solves an issue that we can easily see with maliit
where it would be left in a half-initialised state and would show
autocorrection fields on terminal applications when unnecessary and
possibly similar problems every now and then.
The current UI allows to remove the last desktop, but:
- If you remove it it will reappear on the next KCM launch
- The Add button does not work then
Prevent removing the last desktop by disabling the action
BUG: 437561
When the user applies a change like adding a desktop
DesktopsModel::syncWithServer is called and m_synchronizing is set to
true. The server then sends some kind of response (e.g. desktopCreated)
and updateModifiedState is called which sets m_synchronizing to false
again.
However if a setting is changed that does not trigger any server change
(e.g. the show OSD setting) m_synchonizing is never set to false again,
causing problems down the line.
Instead of relying on updateModifiedState to reset m_synchronizing track
the syncronizing state by refcounting the DBus calls
BUG: 437466
Not choosing any virtual keyboard is a valid option. In this case do not show the SNI since it isn't very useful then and takes up some space in the system tray.
BUG: 437171
Most distros use either systemd's logind or consolekit for session
management.
The main reason why there's direct session is because up to some point,
kwin had a class called VirtualTerminal which was kind of like a direct
session backend.
The VirtualTerminal was used in the fbdev backend in conjuction with the
logind session backend, which looked odd.
Since the drm backend supported only logind and consolekit for very
long time and we hadn't received any complaints about it, the fact that
direct session is unsupported should not be noticeable.
Given that, this change removes the support for direct session in order
to reduce the amount of platform-specific code and make it easier to
maintain kwin.
Currently the apply button will only be enabled when you press enter or
change focus away from the text field which is somewhat inintuitive.
Instead enable the apply button as soon as the text changed.
BUG: 435014
With high DPI we can't use the image size as a logical size directly.
The other change is use of QPainter::drawImage(QRect target, QImage
image, QRect source)
This slightly confusing method uses logical pixels for the target, but
the source rect is in native pixels of the source.
BUG: 422342
If the egl backend gets deleted this leaves a dangling pointer. In
order to prevent crashes with QPainter or EglStreams also don't try
to create dmabuf textures without a EglGbmBackend.
One can think that Item::scheduleRepaint() will trigger repaint of the
whole item, but it actually only schedules the next frame. If nothing
has changed on the screen, no real painting will be done unless maybe if
the rendering backend needs to repair the back buffer.
They are used only by X11Client, so make X11Client call relevants
methods on the surface item directly instead. In hindsight, it will be a
really good idea to make SurfaceItemX11 and SurfaceItemXwayland(?)
automatically manage the window pixmap. However, it can be done once
an item freezing api is added and we fix the cross-fade animation.
Currently, items depend on scene windows for creating pixmaps, repaint
scheduling, and caching quads.
This change moves repaint scheduling from scene windows to items to
make the scene items depend less on scene windows.
In hindsight, we may clean up the repaint scheduling machinery further
by introducing view objects.
So far we had a composition setup within kwinrc and kscreen. This
produces flickering sometime and makes the state a bit more flimsy.
This patch changes the kwin's behaviour to use the files produced by
kscreen which are anyways available down the line.
THis simplifies our behaviour down to just one format that we write to
and feed from. This also allows us to leverage it further by using this
format for default setups (which consist in the right file in
~/.local/share/kscreen).
With the introduction of scene items, the Scene::Window::bufferShape()
method was removed as it makes no sense on wayland - a window may have
several sub-surfaces, so a single region to indicate the shape of the
window won't work.
SurfaceItem::shape() returns the shape of a surface. On Wayland, it
corresponds to the rect of the wl_surface. On X11, it corresponds to the
client window rect inside the frame window, or custom shape region if
the client has set one.
On the other hand, EffectWindow::shape() wants a completely different
thing. If the window is decorated, it needs to return the rect of the
decoration. Otherwise it has to return the shape region if there's one.
In the future, the EffectWindow::shape() function must be removed as it
doesn't fit the item based design. The main reason why we have it at
all is because the x server doesn't support translucency, setting a
shape region is a (hacky) way to work around that limitation, xeyes is
a notable example.
BUG: 437138
BUG: 435862
One of the scene redesign goals is to make wayland surface items
re-usable. So we have the same rendering path for drag-and-drop icons,
software cursors, and window surfaces.
The biggest issue at the moment is that window pixmaps are tightly
coupled with scene windows.
This change de-couples window pixmaps from scene windows. In order to
achieve that, some architecture changes were made.
The WindowPixmap class was replaced with the SurfacePixmap class. A
surface pixmap is created by a surface item.
Under the hood, a SurfacePixmap will create a PlatformSurfaceTexture
object, which contains all the information necessary for the renderer.
The SceneOpenGLTexture class was removed. However, the GLX and the EGL
on X11 backends still mess with GLTexture's internals.
The Toplevel::clientContentPos() function is needed to map the surface
local coordinates to the global coordinates. But its name is highly
confusing as there's already a function with a similar name. This change
introduces a helper with a better name to fix the readability issue.
This is to improve code readability and make it easier to differentiate
between methods that are used during interactive move-resize and normal
move-resize methods in the future.
This makes the implementation of the buffer geometry consistent with the
frame geometry and the client geometry and removes a virtual method call
from a few hot paths.
We need to emit the clientFinishUserMovedResized signal to notify
effects such as translucency that the interactive move-resize is
finished. Otherwise, the set() animation won't be cancelled and the
window will get stuck frozen.
BUG: 409376
As is, kwin with the drm backend results in the most pleasant user
experience on Wayland. Given that and the fbdev being about to be
dropped, making libdrm a required dependency seems a reasonable decision.
Many reasons have been brought up over the years for why this effect is
not appropriate to be enabled by default:
- It was designed to highlight the whizzy technical feature of being
able to make windows transparent, which is no longer particularly
impressive today.
- It looks visually dated.
- It can produce a confusing visual soup by blending a window being
moved or resized with the content below it, which we recently disabled
for the Highlight Windows effect.
- If one window is covering up another as a way to deliberately hide the
content of the lower window, this effect will reveal the hidden
content whenever the upper window is moved or resized at all.
Overall it does not seem to have enough advantages to offset these
issues. Let's disable it by default--but only for the people who did not
modify its configuration at all. We can assume that those people like
it, so let's keep it on for them.
BUG: 384054
FIXED-IN: 5.23
Closes Phab task T7915
It's needed to make kwin_wayland properly systemd managed. A service
launched using systemctl --user is per user, not session.
As a consequence, user services run outside the session. Applications
that deal with user session need some quirks to work with systemd boot.
Unfortunately, GetSessionByPID() doesn't work well with systemd boot,
see bug 433364. GetSession(auto) is shown to work better.
With the client-side decoration changes, kwin will properly determine
whether the window needs to be configured even if the frame geometry has
not changed.
This change slightly changes the semantics of the setFrameGeometry()
method. Prior to this, it was possible to force a geometry, i.e. block
other geometry updates, however such a behavior is counter-intuitive and
it exponentially increases the complexity of code.
As far as I know, the force flag was needed to propagate geometry
changes if the frame geometry doesn't change, but the client geometry
does. With the client-side decoration changes, the force flag is not
needed, as kwin now takes into account the client geometry and the frame
geometry when determining whether to send a configure event.
If a minimized window is animated by the AnimationEffect, the final
layer repaint will be ineffective. In order to account for that, the
AnimationEffect adds the layer repaint to data.paint in the
prePaintScreen() hook. The issue with this hack is that it does not
work on Wayland with per screen rendering.
If a window is animated on screen B, but the layer repaint is added
to data.paint when screen A is painted, there will be a ghost of the
window on screen B.
In order to fix this issue, we need to use effects->addRepaint(), but
we cannot use it in prePaintScreen() as it won't have immediate effect.
effects->addRepaint() can be called either from window paint hooks
or from the postPaintScreen() hook. This change goes after the latter
option.
In the prePaintScreen() hook, AnimationEffect will only advance
animations. In the postPaintScreen() function, it will perform final
janitorial chore.
The internal EventQueue is a child of the registry object. This means
that after the registry is destroyed, all proxy objects in that event
queue are going to have invalid reference to it, which is not a problem
as long as the wl_display_dispatch() function is not called.
The wl_display_dispatch() function uses wl_proxy's queue reference to
enqueue incoming events to that queue.
Unfortunately, during teardown, the internal ConnectionThread may
dispatch events right after the registry object has been destroyed,
which can lead to a crash.
In order to fix the crash, we need to destroy all proxy objects and only
after that we can destroy the event queue. It's okay if wayland events
are dispatched in between.
This allows Task Switcher visualizations to know whether compositing is
active or not so they can adjust their visual representations
accordingly, if needed.
This reverts commit 110182b213.
The reverted commit has introduced a regression where interactively
moved windows jump back to their old position. This needs some time for
proper investigation. Due to beta rolling out quite soon, it will be
better to revert the change.
Rather than have an error prone way to detect whether a configure event
has good position information, handle the case where the plasma shell
surface has a position hint explicitly.
On X11, the lockscreen greeter is an override-redirect window so the
scale and the glide effect ignore it.
On Wayland, the lockscreen greeter is a regular window so both effects
try to animate it upon the screen being unlocked, which looks bad.
So far we were only tracking the first touch point, meaning that if 2
touch points were applied to the internal window and removed one, the
second's events would then be sent to Wayland, which would assert since
it didn't know it had been pressed in the first place.
This fixes InternalWindowTest::testTouch
We want panels to be behind fullscreen windows, but osd windows above
them. With the current layer promotion logic, the panel will be promoted
to the osd layer, which is undesirable. This change restores the old
behavior by allowing layer promotion only to ActiveLayer.
BUG: 436985
It's inherited by XdgToplevelConfigure and we need to make sure that the
object gets destroyed properly.
Alternatively we get a new-delete-type-mismatch error in ASAN and
rightfully so.
Workspace::constrainedStackingOrder() is a constant source of issues.
This change replaces the current logic that builds the stacking order to
hopefully make constrainedStackingOrder() less crashy.
With the new version, constrainedStackingOrder() doesn't perform layer
caching. On the other hand, it's quite rare to see apps in the wild that
have more than just 1 or 2 members in the window group.
BUG: 436095
Currently, the Workspace has no any api to constrain one window above
another. This results in having hacks such as keepDeletedTransientAbove()
This change introduces a basic api to constrain a given window above
another. It can be used for ensuring that transient windows are placed
above their parents. It also can be used for stacking the outline window
below the move-resize window.
Internal windows may also have transient parents. Because of that, this
change makes the workspace add internal clients to the stacking order by
default. The good thing about it is that it allows us unify some input
related code for "external" windows and internal windows.
According to the HIG, the InCubic easing curve should be used if the
animated item changes its state from visible to invisible and the
OutCubic easing curve if vice versa.
This feature was implemented in commit a66eb1a5b9 earlier
Double tap wake up is not a feature to be implemented at compositor
level but rather at the hardware/kernel level. Double tap timer here
means when screen is turned off, libinput will continue to poll the
touchscreen for new events.
double-tap-to-wakeup is generally interrupt at driver/hardware level
which have ability to wake system up even from the sleep.
Provide a option to disable the double tap timer on kwin side for
devices which makes use of suspend since when in suspend kwin can not
wake device up, and that provides confusing user experience.
This micro optimization is simply not worth having. raiseOrLowerClient()
is not in any hot path, besides that raiseClient() is not the only place
where a window can be raised above other windows. In addition to that,
the most_recently_raised check in raiseOrLowerClient() doesn't take into
account whether the cached window is on the current activity.
SurfaceItemInternal will discard the current pixmap if the buffer size
changes. Also, one of the goal of the scene redesign is to keep things
such as this out of Toplevel subclasses.
When the maximize mode changes from MaximizeFull to either
MaximizeHorizontal or MaximizeVertical, (max_mode & MaximizeFull) will
evaluate to true because MaximizeFull is defined as bitwise OR between
MaximizeHorizontal and MaximizeVertical.
We'll hardly port away from it if it's just on the documentation. At the
moment they're used all over the place.
If they didn't have a replacement, they should be undepreprecated.
Currently, the fullscreen state is update synchronously, but it needs to
be done in asynchronous fashion.
This change removes some tests as they don't add any value, testFullscreen()
covers them all.
